metadata

license: mit
library_name: peft
tags:
  - alignment-handbook
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
base_model: microsoft/phi-2
model-index:
  - name: phi-2-dpo-ultrachat-lora
    results: []

phi-2-dpo-ultrachat-lora

This model is a fine-tuned version of lole25/phi-2-sft-ultrachat-lora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.6912
Rewards/chosen: -0.0072
Rewards/rejected: -0.0111
Rewards/accuracies: 0.3180
Rewards/margins: 0.0040
Logps/rejected: -95.3090
Logps/chosen: -92.4438
Logits/rejected: 0.8021
Logits/chosen: 0.7828

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-06
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 4
total_train_batch_size: 64
total_eval_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.693	0.21	100	0.6931	-0.0005	-0.0008	0.2680	0.0004	-94.2804	-91.7748	0.8176	0.7998
0.6922	0.42	200	0.6924	-0.0018	-0.0032	0.3020	0.0014	-94.5141	-91.9068	0.8121	0.7941
0.6917	0.63	300	0.6917	-0.0049	-0.0077	0.3100	0.0028	-94.9659	-92.2189	0.8057	0.7870
0.6905	0.84	400	0.6913	-0.0070	-0.0105	0.3280	0.0036	-95.2509	-92.4247	0.8012	0.7827

Framework versions

PEFT 0.7.1
Transformers 4.36.2
Pytorch 2.1.2+cu118
Datasets 2.14.6
Tokenizers 0.15.2